vous êtes sur le clearnet. les adresses listées ici ne s'ouvrent qu'à l'intérieur du réseau tor - téléchargez le navigateur tor ici »
AlphaBay.Market
FR
dernière vérification : il y a 0 min 255 onions suivis
accueil / actus / réseau tor
03 décembre 2024 réseau tor 4 min de lecture

Le débat sur les onion services à un saut : vitesse contre protection

Every onion service is designed to hide both parties, not just the visitor. A single configuration flag can quietly drop the server side of that promise. Single-hop mode is Tor's official compromise between circuit speed and server anonymity, and it remains one of the network's most misunderstood options.

What one hop actually removes

A standard onion connection stitches together two three-hop circuits, one built by the client and one by the service, joined at a rendezvous point. Six relays stand between the two ends. That redundancy is exactly why connections feel sluggish compared with the clearnet. Single onion services keep the client's circuit untouched but replace the server-side paths with direct, one-hop circuits to the introduction and rendezvous points. The total path shrinks from six hops to four. Connection setup gets faster, streams feel snappier, and the service consumes far less relay capacity. Crucially, the cryptography itself does not weaken. The .onion address still proves the server controls the private key, and traffic remains encrypted end to end. As the rendezvous single onion specification notes, clients cannot even tell whether they are talking to a single or double onion service unless the site tells them.

Why the speed argument is real

The performance case is documented, not hypothetical. Proposal 252, which sketched the original design, lists much lower connection latency, reduced stream latency, and improved scalability as core motivations. Skipping rendezvous circuits also frees busy sites from building and rebuilding long paths for every visitor. Measurements back this up. In a 2020 academic study of exit-blocking circumvention, researchers found that running a bridge as a single onion service added a median of just 0.08 seconds of latency, versus 0.25 seconds for a conventional hidden service. For interactive applications, that difference is plainly user-visible. That finding comes from the HebTor paper, and it captures the appeal. Operators whose servers were never meant to be location-hidden anyway face a fair question: why pay an anonymity tax when the address behind the service is already public knowledge?

What the service gives up

"Single onion services are a modified form of onion services, which trade service-side location privacy for improved performance, reliability, and scalability." - Tor Proposal 252
The trade is blunt. The operator's IP address becomes directly visible to introduction points, rendezvous points, and anyone observing the network around the server. One compromised or hostile relay in those positions learns exactly where the site lives. For whistleblower platforms or anything facing takedown pressure, that alone is disqualifying. Visitors lose something subtler too. Single onions produce distinctive four-hop paths, so traffic analysis can statistically distinguish visits to them, thinning the cover that protects everyone else. The proposal authors themselves warned that each additional flavour of onion service splits the anonymity set shielding ordinary hidden services.
  • The server's real IP address is exposed to intro and rendezvous relays
  • Four-hop paths can be fingerprinted, shrinking the shared anonymity pool
  • Clients have no way to verify which mode a service is running
  • A simple config mistake silently turns an anonymous service public

Built-in guardrails against mistakes

Tor's developers clearly expected carelessness. Enabling single-hop requires setting HiddenServiceSingleHopMode together with HiddenServiceNonAnonymousMode, and the process refuses to start if existing service keys look anonymous. New keys are "poisoned" with a marker file so the same directory can never later be relaunched as a location-hidden service. Further rules close side channels. An instance running a single onion cannot simultaneously act as an anonymous SOCKS client, because traffic could leak across modes, and entry guards are switched off since one-hop circuits render them meaningless. The implementation history shows a design that assumes operators will err, then makes the error loud instead of fatal.

A tool, not a default

Single-hop mode is best read as honest engineering. It concedes that some sites want self-authenticating addresses and client protection without paying for server hiding, and it prices that concession transparently. Used knowingly, it is a sound trade. Used carelessly, it deletes half of a threat model. For visitors, the practical advice stands regardless of how a site is configured. Keep relying on your own three hops, assume nothing about the backend, and read our how onion services work primer before trusting any endpoint. Ongoing coverage lives in our tor network notes.

plus de notes

toutes les news ›
03 sept. 2026 réseau tor lire ›

L’UE rétablit Chat Control 1.0 : le scan des messages revient malgré un vote majoritaire contre

Le Parlement européen n’est pas parvenu à bloquer la relance du régime temporaire et volontaire de scan des messages le 9 juillet 2026 - 314 députés s’y sont opposés, mais le seuil de la majorité absolue n’a jamais été atteint. Alors que les négociations sur le CSAR permanent reprennent en septembre, la bataille du scan et du chiffrement façonne directement l’avenir des outils d’anonymat.

01 sept. 2026 réseau tor lire ›

Telegram veut .gram : que se passe-t-il quand l’infrastructure criminelle quitte les onions pour un DNS contrôlé par la plateforme

Telegram a demandé à l’ICANN le domaine de premier niveau .gram, transformant les noms d’utilisateur en domaines résolvables avec des sites générés par IA derrière. Quelques semaines après la suspension de t.me sous pression des sanctions, le geste montre comment l’infrastructure s’éloigne des onion services vers un DNS que les régulateurs peuvent réellement atteindre.